Model optimization algorithm for target detection YOLOv3 based on deep learning

Abstract

The invention discloses a model optimization algorithm for target detection YOLOv3 based on deep learning. The model optimization algorithm comprises the following steps: resetting Anchor box of an appropriate commodity data set by adopting a K-means++ clustering algorithm; carrying out general training and sparse training on the model of the target detection YOLOv3; taking a final model after YOLOv3 sparseness as a reference, using channel pruning and layer pruning in an overlapping manner to perform double pruning, and pruning unimportant feature channels and layers; and finely adjusting thepruned model, taking a value with a better effect according to the mAP curve graph, and evaluating the obtained value again. According to the target detection YOLOv3 model optimization algorithm based on deep learning provided by the invention, the clustering effect of the algorithm is improved through K-means++; the double pruning combining the layer pruning and the channel pruning is adopted toperform network pruning so as to improve the performance of the algorithm.

Description

technical field [0001] The present invention relates to the technical field of target detection algorithms, in particular to a model optimization algorithm for target detection YOLOv3 based on deep learning. Background technique [0002] The traditional target detection algorithm can be divided into three steps: area selection, feature extraction, and classifier classification. Feature extraction is performed by manually selecting image features, and the features are single and robust. The emergence of convolutional neural network has changed this situation. It can not rely on artificial feature extraction. It has been flourishing after a major breakthrough in the field of image classification. At present, the target detection algorithm based on deep learning has become the mainstream of target detection research. [0003] In recent years, the birth of deep learning target detection algorithms such as regression-based YOLO and SSD has quickly occupied the "market" of target ...

AI Technical Summary

Problems solved by technology

[0004] First, the model is too large, and the powerful representation ability of CNN comes from its millions of trainable parameters

These parameters and network structure information need to be stored on disk and loaded into memory during inference. For example, storing the YOLOv3 model trained on the COCO dataset will consume more than 200MB of space, which is a large resource burden on the device. Especially mobile devices such as embedded;

[0005] Second, it takes up too much memory at runtime. In forward reasoning, the activation layer of CNN may take up more memory space than storing model parameters. Unlike GPU, CPU devices with low computing power are overburdened;

Images